You are on page 1of 8

Volume 12 • Number 1 • 2009


An Empiric Estimate of the Value of Life: Updating the Renal

Dialysis Cost-Effectiveness Standard

Chris P. Lee, PhD,1 Glenn M. Chertow, MD, MPH,2 Stefanos A. Zenios, PhD3
The Wharton School, University of Pennsylvania, Philadelphia, PA, USA; 2Division of Nephrology, Department of Medicine, Stanford
University, Stanford, CA, USA; 3Graduate School of Business, Stanford University, Stanford, CA, USA


Objectives: Proposals to make decisions about coverage of new technol- Results: The incremental cost-effectiveness ratio of dialysis of current
ogy by comparing the technology’s incremental cost-effectiveness with practice relative to the next least costly alternative is on average $129,090
the traditional benchmark of dialysis imply that the incremental cost- per quality-adjusted life-year (QALY) ($61,294 per year), but its distribu-
effectiveness ratio of dialysis is seen a proxy for the value of a statistical tion within the population is wide; the interquartile range is $71,890 per
year of life. The frequently used ratio for dialysis has, however, not been QALY, while the 1st and 99th percentiles are $65,496 and $488,360 per
updated to reflect more recently available data on dialysis. QALY, respectively. Higher incremental cost-effectiveness ratios were asso-
Methods: We developed a computer simulation model for the end-stage ciated with older age and more comorbid conditions. Sensitivity to model
renal disease population and compared cost, life expectancy, and quality- parameters was comparatively small, with most of the scenarios leading to
adjusted life expectancy of current dialysis practice relative to three less a change of less than 10% in the ratio.
costly alternatives and to no dialysis. We estimated incremental cost- Conclusions: The value of a statistical year of life implied by dialysis
effectiveness ratios for these alternatives relative to the next least costly practice currently averages $129,090 per QALY ($61,294 per year), but is
alternative and no dialysis and analyzed the population distribution of the distributed widely within the dialysis population. The spread suggests that
ratios. Model parameters and costs were estimated using data from the coverage decisions using dialysis as the benchmark may need to incorpo-
Medicare population and a large integrated health-care delivery system rate percentile values (which are higher than the average) to be consistent
between 1996 and 2003. The sensitivity of results to model assumptions with the Rawlsian principles of justice of preserving the rights and interests
was tested using 38 scenarios of one-way sensitivity analysis, where of society’s most vulnerable patient groups.
parameters informing the cost, utility, mortality and morbidity, etc. com- Keywords: computer simulation, cost-effectiveness analysis, quality-
ponents of the model were by perturbed +/-50%. adjusted life-years, renal dysfunction, willingness-to-pay.

Introduction cost-effectiveness of dialysis or examined the implications of

adopting a dialysis-related threshold as the basis for coverage
New medical technologies may improve patient outcomes, but decisions. In particular, the broader policy implications of using
generally contribute to rising health expenditures [1,2]. Existing a threshold calculated based on dialysis practice can be contro-
legislation and conventional medical ethics require managed care versial, because it is implied that this threshold is a good proxy
organizations and public payers to cover new medical technology for the society’s valuation of a statistical year of life.
as long as it is “reasonable and necessary” without consideration The objectives in this study were to use current dialysis prac-
of costs [3]. Although the definition of “reasonable and neces- tice and utility estimates to calculate the cost-effectiveness of
sary” is left ambiguous, the decisions made by payers are gener- dialysis and examine how this ratio varies based on changes in
ally based on the strength of clinical evidence supporting the new practice patterns (especially the timing of initiation of dialysis)
technology, especially when the technology is very expensive [3]. and within patient subgroups. These analyses enabled us to
Nevertheless, recent debate about the cost of the new Medicare develop a range of estimates for the cost-effectiveness of dialysis
prescription drug benefit program (part D) suggests that continu- that could potentially be used as a threshold for coverage deci-
ing on the path where coverage decisions are based on clinical sions. The process also enabled us to demonstrate how the data
evidence alone without consideration of costs may not be feasible utilized in the analysis can be used to estimate the value of life
in the long-run. The impending change in legislation has led and to examine the implications of this estimate on future cov-
several researchers to argue that coverage decisions should be erage decisions for expensive medical technologies.
based on both cost and effectiveness criteria, where new technol-
ogy with cost-effectiveness ratios below $50,000 to $100,000 per
incremental quality-adjusted life-year (QALY) is deemed suitable Methods
for coverage, while others with higher ratios are too expensive
Study Design
[4]. The “threshold” of $50,000 to $100,000 is frequently jus-
tified based on the cost-effectiveness of dialysis—an admittedly We developed a computer simulation model for the end-stage
expensive but effective technology that seems to define the renal disease (ESRD) population to examine the incremental
boundary of the highest dollar amount to be paid for an improve- cost-effectiveness ratio (ICER) of dialysis relative to a variety of
ment in QALYs [5]. Yet, no recent studies have established the alternatives, including no dialysis and delayed dialysis. The out-
comes for the model included: life expectancy, life expectancy
Address correspondence to: Chris P. Lee, The Wharton School, University adjusted for quality of life, and economic costs (total societal
of Pennsylvania, 3730 Walnut Street, Suite 500, Philadelphia, PA 19104, costs in 2003 US$) discounted at a 3% annual rate and repre-
USA. E-mail: sented in net present values. ICER was calculated relative to the
10.1111/j.1524-4733.2008.00401.x hypothetical reference cases of no dialysis or delayed initiation.

80 © 2008, International Society for Pharmacoeconomics and Outcomes Research (ISPOR) 1098-3015/09/80 80–87
Value of Life 81

The sensitivity of the results to model assumptions was tested dialysis until eGFR fell a further 1.5 mL/min/1.73 m2 plus an
using 38 scenarios of one-way sensitivity analysis to be described additional 0.1, 0.4, or 0.7 mL/min/1.73 m2 for each 1 point of
in greater detail below. Charlson morbidity score below 10. Allowing initiation to
depend on the Charlson score (higher for patients suffering from
Data Sources greater morbidity) essentially means that healthier patients
would start dialysis later. The delay strategies were specified to
Model parameters were estimated from the following data
mimic a clinically plausible range of delays under resource con-
sources: The United States Renal Data System (USRDS) provided
straints to enable the measurement of the medical “value” of
data on outcomes and costs from more than 500,000 patients
renal dialysis; this is explained in greater depth in the Discussion
initiating dialysis between 1996 and 2003, as well as from
[12]. We obtained an ICER for each of the strategies by dividing
159,616 patients who received a transplant during the same
the difference in cost by the difference in QALYs between that
period [6]. Kaiser Permanenente Northern California provided
strategy and the next least costly strategy [13]. The largest of the
data on disease progression from more than 1.1 million patients
ICERs thus obtained provided an estimate for the cost-
with reduced kidney function cared for between 1996 and 2002.
effectiveness threshold implied by current dialysis practice and
Quality of life data were obtained from direct inquiries of utility
the implied value of life (more on this in the Discussion).
[7] that differ by kidney function (as measured by the estimated
In addition, we determined whether this threshold of value of
glomerular filtration rate, eGFR), as well as whether the patient
life would change depending on subgroup analysis. A sample of
required dialysis. Two sets of utility estimates were provided
1000 patients was simulated 10,000 times to calculate the ICER
using alternative methods of quality adjustment: the time
for current practice relative to the next least costly strategy for
tradeoff (TTO) and Health Utilities Index Mark 3 (HUI-3) [8].
each patient in the cohort. The cohort of 1000 patients was then
The midpoint of the estimates was used as baseline. We consid-
divided into quintiles of cost-effectiveness, and both the median
ered perturbations to the high end (TTO) and low end (HUI-3) in
and average ICERs within each quintile were computed. To
the sensitivity analysis.
determine the relations among demographic factors, comorbid
conditions, and cost-effectiveness, we evaluated the distribution
The Simulation Model of these factors across quintiles of cost-effectiveness. The latter
Details of the simulation model have been previously reported [9]. evaluation was performed to determine the population range of
A patient generation model generated a cohort of 1,000,000 cost-effectiveness and the potential role of patient characteristics
patients, and a patient simulation model evolved the profile of in determining the difference.
each patient over time. The Patient Generation model generated For comparison, we repeated the analysis using less sophisti-
random patient profiles for each patient in the cohort. The profile cated strategies, where the delay was uniform across all patients
included each patient’s age, sex, race and ethnicity, blood type (for irrespective of the underlying severity by 12, 24, or 48 months.
simulating the time to transplantation), comorbidities (diabetes,
atherosclerotic cardiovascular disease, congestive heart failure,
and cancer), eGFR, and serum albumin. Each patient’s profile was Sensitivity Analysis
generated by sampling from the empirical distribution of the Baseline parameter values of the simulation model were reported
incident patient population of the USRDS. in [9]. There are over 130 parameters describing various costs,
The Patient Simulation model generated a medical history for utility, and hazard submodels within the simulation model. A
each patient by simulating the time between the following events 38-scenario sensitivity analysis was conducted, where in each
that modified the patient’s profile, determined costs, and affected scenario, one or a related group of parameters would be per-
quality of life and survival: 1) eGFR deterioration capturing the turbed by +/-50% from their baseline values, and the ICER
gradual loss of kidney function; 2) hospitalization to capture would be recalculated based on new simulation results. The
hospital inpatient episodes; 3) transplantation to capture patients purpose was to evaluate the uncertainty in which parameters
receiving a transplant; 4) graft failure indicating return to dialy- might affect the ratio and by how much. The breakdown of the
sis; and 5) death. The time between events is assumed to be scenarios are: two scenarios for the rate of hospitalization, two
exponentially distributed time inhomogeneously, and the mean
time between events is modified dynamically by changes in
patients’ attributes. A summary of the cost parameters and utility
Table 1 Model assumptions about costs and utilities
scores is provided in Table 1 [10,11]. All remaining parameters
are summarized in [9]. Estimate Source
Because the main objective of the study was to estimate the
cost-effectiveness threshold implied by current dialysis practice Cost ($)
Hospitalization 12,831 (6,416, 19,247)
and extrapolate from it an estimate for the value of life, we Transplantation 81,330 (40,665, 121,995)
simulated a dialysis strategy that reflected current practice, and Transplant follow-up 15,735 (7,868, 23,603) Medicare
additional three strategies where patients would start dialysis Graft failure 29,392 (14,696, 44,088)
later than in current practice, i.e., strategies that are likely to be Dialysis, fixed + EPO 153 (77, 230)
Dialysis, per minute 0.42 (0.21, 0.63)
less costly in terms of (remaining) lifetime cost per patient, Off-transplant quality of life
because patients would now live shorter and spend less time on 15 ⱕ GFR ⱕ 30 0.700 (0.35, 1.05) Gorodetskaya
dialysis, although only simulation would confirm this because et al. [7]
there would also be a rise in hospital costs. In the Current GFR < 15, no dialysis 0.695 (0.348, 1.043)
Dialysis 0.630 (0.315, 0.945)
Practice strategy, dialysis was started according to a regression On-transplant quality of life 0.825 (0.413, 1.238) Laupacis et al. [10],
function capturing the common practice of starting dialysis Hornberger
roughly when a patient’s eGFR dropped below 9 mL/min/ et al. [11]
1.73 m2. In the three delay strategies—Current Practice with Discount rate 0.03 (0.015, 0.045) —
Slight Delay, Current Practice with Moderate Delay, and Current Baseline values are followed by perturbation limits used in the sensitivity analysis.
Practice with Significant Delay—patients would not be started EPO, erythropoietin; GFR, glomerular filtration rate.
82 Lee et al.

Table 2 Summary of scenarios for the sensitivity analysis

Scenario no. Scenario summary % change from baseline Sign

0 Baseline
1 Transplant rates up 5 -
2 Transplant rates down 10 +
3 Mean time to hospitalization decreased 2 -
4 Mean time to hospitalization increased 2 -
5 Mortality rates up 0 -
6 Mortality rates down 6 -
7 More graft failures 4 +
8 Fewer graft failures 5 -
9 Rapid eGFR decline 2 -
10 Slow eGFR decline 0 +
11 High discount rate 3 +
12 Low discount rate 8 -
13 Dialysis costs up 35 +
14 Dialysis costs down 39 -
15 Cost of hospitalization up 5 -
16 Cost of hospitalization down 3 +
17 Costs of transplant up 6 +
18 Costs of transplant down 4 -
19 On-dialysis utility up 0 -
20 On-dialysis utility down 2 +
21 On-transplant utility up 1 -
22 On-transplant utility down 3 -
23 Effect of combined dialyzed and native clearance on hospitalization amplified 5 -
24 Effect of combined dialyzed and native clearance on hospitalization deamplified 31 +
25 Effect of dialysis frequency on hospitalization amplified 8 -
26 Effect of dialysis frequency on hospitalization deamplified 11 +
27 Effect of dialysis duration on hospitalization amplified 7 -
28 Effect of dialysis duration on hospitalization deamplified 1 -
29 Effect of zero clearance on hospitalization amplified 4 -
30 Effect of zero clearance on hospitalization deamplified 21 +
31 Effect of combined dialyzed and native clearance on mortality amplified 8 -
32 Effect of combined dialyzed and native clearance on mortality deamplified 27 +
33 Effect of dialysis frequency on mortality amplified 0 -
34 Effect of dialysis frequency on mortality deamplified 3 +
35 Effect of dialysis duration on mortality amplified 7 -
36 Effect of dialysis duration on mortality deamplified 4 +
37 Effect of zero clearance on mortality amplified 2 -
38 Effect of zero clearance on mortality deamplified 4 +

eGFR, estimated glomerular filtration rate.

scenarios for the rate of mortality, two scenarios for the rate of Results
eGFR decline, two scenarios for the rate of transplantation, two
scenario for the rate of graft failure, eight scenarios for the effect Cost-Effectiveness of Dialysis for End-Stage
of dialysis on mortality, eight scenarios for the effect of dialysis Renal Disease
on hospitalization, four scenarios for health utility, two scenarios Table 3 presents the outcomes for the different dialysis strate-
for the costs of dialysis, two scenario for the cost of hospital gies considered. Relative to No Dialysis, dialysis increased
admissions, two scenarios for the costs of transplantation, patient life expectancy by an average of 34.11 months (Current
follow-up, and graft failure, and two scenarios for the discount Practice). The total lifetime costs increased from $135,076 (No
rate. Baseline values and perturbation limits for some of the Dialysis) to $281,640 (Current Practice), respectively. The
parameters are shown in Table 1, while the scenarios are sum- ICERs, which in Table 3 are calculated relative to the next least
marized in Table 2. costly strategy, depended on the timing of initiation: Current

Table 3 Cost-effectiveness of different dialysis strategies

No. of Current practice with Current practice Current practice Current

Outcome dialysis significant delay with moderate delay with slight delay practice

Mean survival (months) 47.88 50.78 58.05 68.99 81.99

Mean quality-adjusted survival (quality-adjusted life-months) 28.68 30.32 33.68 38.38 44.55
Mean lifetime cost per person ($) 135,076 140,590 168,820 215,260 281,640
Incremental cost-effectiveness ($ per quality-adjusted life-year 40,446 100,717 118,540 129,090
Incremental cost-effectiveness ($ per life-year gained) 22,792 46,594 50,938 61,294
Mean delay by time (months) 24.29 18.69 11.05
Mean delay by eGFR (ml/minutes/1.73 m2) 5.75 4.27 2.62

Each incremental cost-effectiveness ratio is calculated by dividing the cost difference between a strategy and the strategy to its immediate left by the survival difference (in either years or
quality-adjusted years) between the same two strategies.
Slight Delay, Moderate Delay, and Significant Delay mean dialysis would not start until (relative to Current Practice) eGFR fell a further 1.5 ml/min/1.73 m2 plus an additional 0.1, 0.4, or
0.7 ml/min/1.73 m2 for each 1 point of Charlson morbidity score below 10.
eGFR, estimated glomerular filtration rate.
Value of Life 83

Table 4 Incremental cost-effectiveness between different pairs of strategies

Cost-effectiveness No Current practice with Relative to: current practice Current practice Current
ratio ($/QALY) for dialysis significant delay with moderate delay with slight delay practice

No dialysis 110,814
Current practice with significant delay 40,446 129,090
Current practice with moderate delay 80,993 100,717 124,528
Current practice with slight delay 99,189 118,540 118,902
Current practice 110,814 129,090 —

QALY, quality-adjusted life-year.

Practice with Significant Delay had the lowest ICER of $40,446 be used to update the frequently cited range of $50,000 to
per QALY, which increased to $129,090 per QALY with $100,000 per QALY for the incremental cost-effectiveness of
Current Practice. When expressed in costs per life-years gained dialysis.
(as opposed to quality adjusted life-years), the ratios ranged
from $22,792 to $61,294.
ICERs can also be calculated relative to No Dialysis. Table 4 Patient Characteristics and Cost-Effectiveness
provides ratios calculated between different pairs of strategies— of Dialysis
namely, between each strategy and No Dialysis (first column), The cost-effectiveness of dialysis differed among population sub-
between each strategy and Current Practice (last column), and groups. Figure 1 displays the incremental cost-effectiveness of
between each strategy and the next least costly strategy (diago- Current Practice (relative to Current Practice with Slight Delay)
nal). In this study, we focus on ratios calculated relative to the with patients ranked into five quintiles. The median ICERs
next least costly strategy, and a justification for this approach is ranged from $99,749 per QALY for the first quintile to $240,010
provided in the Discussion. The most notable figures from this per QALY for the fifth quintile. Patients in the quintile with the
table are $110,814 per QALY for Current Practice relative to highest ratios were more likely to be older with more comorbid
No Dialysis, and $129,090 per QALY for Current Practice conditions (data not shown). A distribution of the ICER is pro-
relative to Current Practice with Slight Delay. These figures can vided in Figure 2.

5th Quintile

4th Quintile

3rd Quintile

2nd Quintile

1st Quintile

- 50,000 100,000 150,000 200,000 250,000 300,000

1st Quintile 2nd Quintile 3rd Quintile 4th Quintile 5th Quintile
Median 99,749 133,340 154,840 187,400 240,010
Average 89,150 129,640 153,980 184,330 249,660

Figure 1 Quintiles of the distribution of cost-effectiveness ratios.

84 Lee et al.










0 1 2 3 4 5 6
x 10

Cumulative Distribution: (100-α)%

0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0.0 1,304 65,496 77,616 83,800 86,239 88,924 90,838 92,861 94,101 96,621
0.1 99,749 102,320 104,060 106,170 107,760 109,200 110,870 111,940 113,670 114,770
0.2 116,970 118,850 120,880 122,290 123,560 125,280 127,030 128,540 130,100 132,480
0.3 133,340 134,110 135,910 137,260 138,390 139,330 140,180 141,110 142,440 143,490
0.4 144,260 144,820 146,350 147,390 148,230 149,800 150,730 151,820 153,020 153,880
0.5 154,840 156,330 158,170 160,880 161,460 163,590 165,240 166,610 168,160 169,420
0.6 170,590 172,740 173,960 174,990 176,550 177,920 179,810 181,490 183,350 185,680
0.7 187,400 188,140 190,400 192,700 195,040 197,170 198,810 200,850 202,110 203,570
0.8 206,560 207,910 210,850 212,650 214,470 218,750 223,990 226,730 232,330 235,760
0.9 240,010 248,840 261,410 269,340 290,700 319,110 359,260 466,910 477,630 488,360

Figure 2 Tail distribution of the incremental cost-effectiveness ratio for Current Practice (relative to Current Practice with Slight Delay).

Sensitivity Analysis
The last column of Table 2 provides model sensitivity as mea-
sured by the absolute percentage change in the ICER (Current
Practice relative to Current Practice with Slight Delay) induced
by the +/-50% perturbation in the parameters. The sensitivity 46
was comparatively small, with most scenarios leading to a 44
change of no more than 10%. More significant changes were 42
found in scenarios related to changes in the costs of dialysis (13
and 14, at 35% and 39%, respectively) and in the ability of
dialysis to attenuate hospitalizations (24 and 30, at 31% and
21%, respectively) and mortality (32, at 27%). Measuring sen- 36
sitivity using the ICER relative to No Dialysis did not produce an 34
appreciable difference in the pattern or magnitude of sensitivity. 32
Efficiency Frontier 28
Figure 3 shows that delay strategies based on the Charlson score 135,000 185,000 235,000 285,000 335,000
dominated the simpler strategies of uniformly delaying all Mean Lifetime Cost per Patient ($)
patients by a fixed amount of time. In this article, we focus on
ICERs relative to the former set of strategies the rationale for Figure 3 Cost-effectiveness efficiency frontier. The efficiency frontier was
which is discussed below. obtained by tracing delay strategies based on the Charlson score with a smooth
fit. Strategies based on delaying all patients uniformly are found in the interior
of the frontier. From left to right, the diamonds correspond to: No Dialysis,
Discussion Current Practice with Significant Delay, Current Practice with Moderate Delay,
Current Practice with Slight Delay, and Current Practice; the squares are:
The cost of dialysis per QALY gained is frequently quoted as a Current Practice with Uniform 48-month Delay, Current Practice with Uniform
benchmark for the cost-effectiveness of medical technologies. 24-month Delay, and Current Practice with Uniform 12-month Delay.
Value of Life 85

The most commonly used number is $50,000 per QALY [4], survey of estimates based on occupational risk by Viscusi and
while a more recent study adjusted that number to $93,500 per Aldy found a range from $500,000 to $21 million per statistical
QALY by inflating the earlier number to 2002 US$ [14]. The life [19]. Another approach is based on the cost-effectiveness of
original estimate, which can be traced to a 1984 Canadian study, life-saving interventions in nonmedical fields, such as occupa-
was based on the accounting ledger for 44 dialysis patients at one tional health, transportation safety, or environmental hazard
center during a time span of 1 year and a sophisticated cost- control [15,16]. Estimates using these methods ranged from
allocation algorithm [5]. Our analysis based on a comprehensive $56,000 per life-year saved for transportation programs to $4.2
model of the ESRD population and recent data on cost, utility, million per life-year saved for environmental programs [20].
and disease progression suggests that this benchmark has We should provide some justification for the methodologies
increased beyond the rate of inflation to exceed the $93,500 per of using strategies based on delaying dialysis, as well as our
QALY figure: a more accurate figure is between $110,814 per calculation of ratios by comparing with the next least costly
QALY (when Current Practice is compared with No Dialysis) strategy. In theory, cost-effectiveness analysis (CEA) is a heuris-
and $129,090 per QALY (when Current Practice is compared to tical approximation to an optimal resource allocation problem
Current Practice with Slight Delay). The increase could be [21], and the value of the CEA threshold is endogenously deter-
because of the higher than anticipated pace of health expenditure mined by the exogenous budget. Most actual applications of
inflation (“price”), or to innovations in nephrology and dialysis CEA reverse this process by directly setting the threshold to avoid
care, such as recombinant erythropoietin, available only after the the appearance of explicit budget setting [12]. Given its long and
original 1984 estimate (“treatment mix”). The increase might unique history of Medicare coverage, dialysis is believed to
also be simply because of more widespread use of dialysis than provide a justifiable benchmark for setting the threshold to, i.e.,
before (“technology diffusion”). Alternatively, the utility of it represents socially accepted medical “value” ($/QALY). To
patients on dialysis was estimated to be lower than previously determine the medical value of dialysis, we assume the level at
utilized values [7]. which dialysis is currently utilized (i.e., Current Practice) derives
The frequent use of the cost-effectiveness of dialysis as bench- from a formal decision process. In that decision, physicians could
mark, as well as proposals of using it as the threshold for cov- have chosen less dialysis (as represented by Current Practice with
erage decisions, implies the belief or perception that the cost- Slight Delay), but they did not. That last and most expensive
effectiveness of dialysis reflects the society’s valuation for a increment of dialysis ($129,090 per QALY) must thus define an
statistical year of life. This can be justified based on the economic implicit value threshold. Although ICERs are more frequently
argument that society’s willingness to pay for medical interven- calculated relative to the nonuse of a medical intervention (i.e.,
tions (on a $ per QALY basis) must at least equal the value No Dialysis), we believe calculating ICERs with respect to a
generated by dialysis for the latter to be universally covered by slight delay is reasonable, because the dialysis decision involves a
Medicare, and thus decisions for other medical interventions can continuous “timing” dimension; indeed, the timing of dialysis is
be made relative to dialysis. Indeed, the role of dialysis in the a matter of intense debate in the nephrology community and has
history of Medicare is an important and unique one: because spawned policy discussions and recent changes in guidelines [22].
Medicare initiated its coverage of ESRD in 1973, dialysis (and A final point is related to our use of the Charlson score in
more generally renal replacement therapy) has remained the only determining the amount to delay dialysis. The intention here is to
example where coverage is granted in the United States solely on better capture how decisions to delay might be carried out in
the basis of a diagnosis [5]. There is also the counterargument practice: if forced into the situation of having to delay dialysis
that universal coverage for dialysis is an anomaly and hence (because of capacity or budget reasons, for example), physicians
society’s valuation is lower than that implied by dialysis. Irre- would be most reluctant to do so with sicker patients. As a result,
spectively of one’s position on the relevance of dialysis as a we’d expect the sickest patients (i.e., those with the highest
benchmark, historical precedence suggests that ICER of dialysis Charlson score) to experience the least delay, whereas the healthi-
will either approximate society’s true valuation of life or provide est patients would experience the most delay. Another compelling
a useful bound for it. The estimates derived here are consistent reason is that delaying dialysis by a fixed amount of time uni-
with numbers reported elsewhere: The World Health Organiza- formly across patients is clinically suboptimal. Figure 3 shows
tion proposes $108,600 per disability-adjusted life-year [15]. that the strategies of delaying dialysis uniformly across patients
When the threshold is expressed in dollars per life-year saved, the by 12, 24, and 48 months are in the interior of the efficiency
estimate derived from dialysis is $61,294 per life-year. This is frontier defined by the delay strategies based on the Charlson
comparable with the range of $55,000 to $88,000 (2000 US$) score, i.e., they are dominated. Formally optimizing dialysis
reported in [5]. It is also consistent with the average of $65,000 strategies is beyond the scope of this article and is explored in
per life-year gained obtained from a survey of health economists [23].
and comparable with the range of $44,800 to $83,900 per life- The results in Figure 1 show that when distributional consid-
year gained estimated from cardiovascular interventions [13]. erations are important, making decisions based on the popula-
The value of a statistical year of life can also be estimated tion average of incremental cost-effectiveness ratio as the
from nonclinical data. A common approach is to calculate the threshold can be challenging. Consider what would happen if
relative increase in salary required for a worker to incur an coverage decisions were to be made based on the $129,090 per
increase in occupational risk [16]. This method yielded an esti- QALY ($61,294 per year) figure. A straightforward application
mate of $428,286 per QALY in a recent study [17]. Salaries of this number would imply that a technology would be covered
offered to contractors in Iraq, ranging from $60,000 to $175,000 for a population as long as its average cost-effectiveness would be
per year, reflect a modern example of the willingness of persons below this figure. Nevertheless, it is possible that this population
to make economic choices where risks are palpably increased can be divided into subgroups, with the cost-effectiveness in some
[18]. Assuming an annual risk of death of 0.004 and a salary subgroups exceeding the threshold, and with some, well below.
premium of $30,000 per year over comparable jobs in the United Providing coverage to groups with cost-effectiveness ratios below
States, and assuming also that dying in Iraq reduces the life the threshold while withholding coverage for the remaining
expectancy by 30 years, contractors in Iraq are essentially com- groups would perhaps not pass most tests of equity and not
pensated at a rate of $250,000 per statistical year of life. A recent represent a sound application of the threshold. This raises the
86 Lee et al.

question of whether there is a better threshold than $129,090 per The analysis has several limitations. The model does not
QALY. One could even argue that a single “best” threshold that capture all comorbidities and cannot describe physiological
works with all applications is elusive and that differential thresh- effects or psychosocial factors that might influence the outcomes
olds may be needed for different subgroups. Just as there are of care. Using simulation, we cannot determine the exact QALYs
various notions of equity—some of which work better than and costs. We tried to control for the margin of error in the
others in certain circumstances, the choice of threshold must estimates of QALYs and costs by simulating large populations. In
begin with a well-defined notion of equity. doing so, we were able to reduce the standard error of the
One way of deriving an equitable threshold is to start with the estimates to <0.005 QALYs and <$300 lifetime costs.
Rawlsian principle of justice: resources should be allocated to Arguably, there are more important limitations related to the
benefit everyone, including the most vulnerable individuals [24]. conclusions that we have drawn from our work. That dialysis
Then, from Figure 2, we can obtain a probabilistic view of the reflects an accepted social willingness to pay depends on the
Rawlsian principle: the (100-a)-th percentile of the incremental assumption that the public and policymakers still believe that
cost-effectiveness ratio for dialysis can be used as a threshold dialysis should be provided to all Americans who require it. The
(i.e., a = 5, 10, or 15), with the recognition that this the least of decision to provide dialysis as a covered benefit under the Medi-
what the threshold needs to be to ensure comparable coverage care Program in July 1973 was made on the basis of medical
for the medical intervention in question as for the most expensive justification and political will, with the expectation that many
a-percent of dialysis patients. Although in a strict sense, the beneficiaries would regain health and return to the workforce
Rawlsian principles would ask that a be set at 0 (i.e. for the [25,26]. Although coverage decisions for new technologies are
society’s least fortunate), doing so would not be financially fea- made with regularity, the decision to approve a new technology
sible in practice. The laws of randomness would ensure that one has different implications than a decision to withdraw a technol-
can always find some person who observes an arbitrarily poor ogy after the latter has been approved and available for decades.
(high) incremental cost-effectiveness ratio, as is shown in the Thus, it is possible that the $129,090 per QALY ($61,294 per
right asymptote of Figure 2. A reasonable way to “amend” the year) figure for the value of life that we have calculated overes-
Rawlsian notion of justice in a world of stochasticity is, there- timates the marginal value that might be determined using assess-
fore, to think of the incremental cost-effectiveness ratio in terms ments based on other, newer technologies.
of the tail distribution. In short, the wide distribution of ICER All of the methods for deriving a threshold presented here are
obtained in our analysis shows that there is no single appropriate based on directly calculating the ICER. A limitation of this
threshold, but rather a continuum of thresholds, which might be approach is that there are many ways to calculating the ICER (as
conceptually thought of as a function in a. Thresholds corre- we showed in Table 4). Depending on the point of reference
sponding to smaller values of a reflect a higher emphasis on the chosen, the results generally differ. This reflects the roots of CEA
distributive implications of the Rawlsian ideal, but also generate as an approximation to an underlying optimization problem:
a greater financial burden. The decision-maker is confronted there are many ways to construct an approximation. Alterna-
with having to make a tradeoff. tively, a threshold can also be derived directly from the underly-
One would be naive to expect that a threshold-based system ing optimization problem. That approach is conceptually more
provide the solution to the problem of increasing health-care sophisticated and requires more technical machinery, but it has
costs. As the dialysis figures indicate, the threshold can creep the appeal of leading to one unambiguous threshold (with a
higher over time, implying that the amount paid for a fixed precise mathematical interpretation). The high-level idea there
increase in quality-adjusted life expectancy may increase over is to set up the current allocation (as represented by Current
time. Nevertheless, it does provide a heuristical framework Practice) as the result of a formal optimization problem with
where both the cost and effectiveness of a technology is assessed, unknown parameters, and iteratively, the unknown parameters
and more importantly, it can provide incentives for innovators to are calculated. One of the parameters is a value threshold, and
develop solutions that might reduce the cost of care without thus it can be viewed as the value threshold implied by the
adversely affecting the quality of care or life expectancy. current practice of dialysis. Another appeal of that approach is
This analysis has several strengths. The USRDS registry that it provides an exploratory framework for assessing the
includes the vast majority (approximately 95%) of all persons degree of inequity currently in the system. A downside is that the
in the United States requiring dialysis and transplantation for approach is computationally intensive (relying on methods of
ESRD, so that the dialysis results described here are generaliz- inverse optimization), which prevents more casual uses. We
able to the US dialysis population. Moreover, we were able to discuss that approach in [27].
simulate changes in practice (e.g., long delays in dialysis initia- In summary, using data from the USRDS and other sources,
tion) that could not be tested in clinical practice. Data from we have determined the cost-effectiveness of dialysis in the
USRDS were supplemented with data from a large integrated modern era. In doing so, based on the assumption that dialysis is
health-care delivery system, incorporating information on hos- a desired benefit to be provided to persons with ESRD, we have
pitalization and outcomes that would otherwise be unavailable determined an empiric value of life. Based on careful simulation,
from the nonelderly and disabled dialysis population without and philosophical principles aimed to protect the vulnerable, we
Medicare as primary payer. We have previously validated the have determined range of dialysis-based thresholds based the tail
simulation model, which yielded results under the Current Prac- distribution. Although no method can definitively determine the
tice strategy virtually identical to empirically observed outcomes actual value an individual places on his or her lifetime, these
[9]. Unlike many other medical technologies recently intro- estimates are less prone to some of the problems faced by esti-
duced, dialysis has a more than 30-year history of use in the mates using labor market data or personal choices involving
United States, and information on costs and outcomes associ- small but finite risks, which have been shown that people tend to
ated with dialysis and transplantation are more granular than overestimate [28]. Whether these estimates will be used to gen-
those available for many other high-cost technologies. The esti- erate policy decisions remains to be determined.
mates of utility were recently obtained, using two conventional,
yet disparate measures, which bracketed the model’s utility The data reported here have been supplied by the USRDS. The interpre-
inputs. tation and reporting of these data are the responsibility of the author(s)
Value of Life 87

and in no way should be seen as an official policy or interpretation of the 13 Garber AM, Weinstein MC, Torrance GW, Kamlet MS. Theoreti-
US government. This study was funded in part by NIH grant NIDDK RO1 cal foundations of cost-effectiveness analysis. In: Gold MR, Siegel
DK58411. JE, Russel LB, Weinstein MC, eds., Cost-Effectiveness in Health
and Medicine. New York: Oxford University Press, 1996.
Source of financial support: This study was supported in part by NIH- 14 Eichler HG, Kong SX, Gerth WC, et al. Use of cost-effectiveness
NIDDK RO1 DK58411 from the National Institute of Health. analysis in health-care resource allocation decision-making: how
are cost-effectiveness thresholds expected to emerge? Value
References Health 2004;7:518–28.
15 WHO Commission on Macroeconomics and Health. Investing in
1 Newhouse JP. Medical care costs: how much welfare loss? J Econ Health for Economic Development. Geneva: World Health Orga-
Perspect 1992;6:3–21. nization, 2001.
2 Cutler DM, McClellan M. Is technological change in medicine 16 Dranove D. What’s Your Life Worth? Health Care Rationing
worth it? Health Aff (Millwood) 2001;20:11–29. . . . Who Lives? Who Dies? And Who Decides? Upper Saddle
3 McClellan MB, Tunis SR. Medicare coverage of ICDs. N Engl J River, NJ: FT Prentice Hall, 2003.
Med 2005;352:222–4. 17 Hirth RA, Chernew ME, Miller E, et al. Willingness to pay for a
4 Gillick MR. Medicare coverage for technological innovations— quality adjusted life year: in search of a standard. Med Decis
time for new criteria? N Engl J Med 2004;350:2199–203. Making 2000;20:332–42.
5 Winkelmayer WC, Weinstein MC, Mittleman MA, et al. Health 18 Available from: [Accessed
economic evaluations: the special case of end-stage renal disease July 19, 2005].
treatment. Med Decis Making 2002;22:417–30. 19 Viscusi WK, Aldy JE. The value of a statistical life: a critical
6 United States Renal Data System. Researcher’s Guide to the review of market estimates throughout the world. J Risk Uncer-
USRDS Database. Bethesda, MD: National Institutes of Health, tain 2003;27:5–76.
National Institute of Diabetes and Digestive and Kidney Diseases, 20 Tengs TO, Adams ME, Pliskin JS, et al. Five hundred life-saving
2004. interventions and their cost-effectiveness. Risk Anal 1993;15:
7 Gorodetskaya I, Zenios SA, McCulloch CE, et al. Health-related 369–90.
quality of life and estimates of utility in chronic kidney disease. 21 Weinstein MC, Zeckhauser RJ. Critical ratios and efficient allo-
Kidney Int 2005;68:2801–8. cation. J Public Econ 1973;2:147–57.
8 Gold MR, Patrick DL, Torrance GW, et al. Identifying and 22 Levey AS, Coresh J, Balk E, et al. National Kidney Foundation
valuing outcomes. In: Gold MR, Siegel JE, Russel LB, Weinstein guidelines for chronice kidney disease evaluation, classification,
MC, eds., Cost-Effectiveness in Health and Medicine. New York: and stratigication. Ann Intern Med 2003;139:137–47.
Oxford University Press, 1996. 23 Lee CP, Chertow GM, Zenios SA. Optimal initiation and man-
9 Lee CP, Chertow GM, Zenios SA. A simulation model to estimate agement of dialysis therapy. Operations Research (in press).
the cost and effectiveness of alternative dialysis initiation strate- 24 Rawls J. A Theory of Justice. Cambridge, MA: Harvard Univer-
gies. Med Decis Making 2006;26:535–49. sity Press, 1971.
10 Laupacis A, Keown P, Pus N, et al. A study of the quality of life 25 Levinsky NG. The organization of medical care—lesson from the
and cost-utility of renal transplantation. Kidney Int 1996;50:235– Medicare end stage renal disease program. N Engl J Med 1993;
42. 329:1395–9.
11 Hornberger J, Redelmeir D, Petersen J. Variability among 26 Levinsky NG. Quality and equity in dialysis and renal transplan-
methods to assess patient’s well-being and consequent cost- tation. N Engl J Med 1999;341:1691–3.
effectiveness analysis. J Clin Epidemiol 1992;45:505–12. 27 Lee CP, Zenios SA. A shadow price framework for quantifying
12 Ledebo I, Kessler M, v Biesen W, et al. Initiation of dialysis— health care demand, spending and disparity. OPIM Department
opinions from an international survey: report on the Dialysis Working paper, The Wharton School, 2008.
Opinion Symposium at the ERA-EDAT Congress, 18 September 28 Johannesson M, Meltzer D. Some reflections on cost-effectiveness
2000, Nice. Nephrol Dial Transplant 2001;16:1132–8. analysis. Health Econ 1998;7:1–7.